Reproduce “One Tile, Multiple Instances: Rethinking MIL for Sparse Diagnostic Evidence” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Runsheng Liu, Cheng Jin, Hao Jiang, Hao Chen
DI-MIL improves weakly supervised whole-slide classification by representing each tile with four independently weighted summaries of dense patch tokens. It clusters normalized tokens by feature similarity, pools original features, and sends them to standard attention aggregators. Across four cytopathology datasets, it improves 67/72 comparisons; on Respiratory, macro-F1 rises from 71.83 to 80.08, but not universally.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.